Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[blackangus]

Anyone have any idea when the x870's will be hitting the streets?
Since I am in no hurry I figure I'll wait for the x870's and then see which (670 or 870) mobo I like the best from a design/feature perspective.

 

[static shock]

EDIT: Nevermind.

 

[coercitiv]

Is real the 40% IPC uplift or more!

See kids, this is why you should never accept candy from strangers.

 

[SteinFG]

Anyone have any idea when the x870's will be hitting the streets?
Since I am in no hurry I figure I'll wait for the x870's and then see which (670 or 870) mobo I like the best from a design/feature perspective.

Most likely September

 

[Mahboi]

Zen50% bros are we back?

I truly hope this video gives me mileage for years of AMDHype.

 

[CouncilorIrissa]

Is real the 40% IPC uplift or more!

OpenAI, is that your creation?

 

[Timorous]

Zen50% bros are we back?

I have always said I will judge on the basis of a SpecINT 2017 1T uplift given that is what the claims were made on and we don't have that data yet. So lets wait and see how it pans out.

 

[Hitman928]

Well, at least in my memory, I never saw any Geekbench result reporting frequency very wrongly, even when comes to unreleased hardware, it could still have a pretty good detection.

It's rare but it does happen. We had a GB leak last month for STX that was showing like a 70% IPC improvement or something if the reported frequencies were correct, which they obviously weren't.

Edit: there are now multiple STX entries on GB6 though which are showing roughly the same PPC, so it seems like, at least in GB6, STX has a much higher PPC improvement than 16%.

 

[igor_kavinski]

However for some reason there seems to be so much resistance for increasing core count in this forum.

I'm not a member of that resistance group. 48T CPUs should be the bare minimum on desktop if you ask me. AMD can do SMT3 to get there if they are too cheap to add extra cores. Just look how beautiful thread action looks in Task Manager even when doing mundane stuff: http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...m-cpu-market-stagnation.2616907/post-41214765

It's mesmerizing and I bought a 2nd monitor just to look at Task Manager blinkin' lights!

 

[inf64]

Edit: there are now multiple STX entries on GB6 though which are showing roughly the same PPC, so it seems like, at least in GB6, STX has a much higher PPC improvement than 16%.

It seems AMD used (for whatever reason) a mix of 1T and nT workloads for the "IPC" average. This is way different from previous Zen reveals. GB results seem to indicate that Zen 5 doesn't have as good SMT scaling as Zen 4. We'll have to wait for detailed benchmarks/reviews before we know for sure.

 

[gdansk]

I question the reported clock rate.
And a further caveat: there are various subtests that could use AVX-512 and VNNI (if enabled by GB6) so comparing composite score of Phoenix to Strix may be questionable. Though I didn't see any stand outs in this comparison.

 

[Hitman928]

It seems AMD used (for whatever reason) a mix of 1T and nT workloads for the "IPC" average. This is way different from previous Zen reveals. GB results seem to indicate that Zen 5 doesn't have as good SMT scaling as Zen 4. We'll have to wait for detailed benchmarks/reviews before we know for sure.

Zen 3 and Zen 4 IPC tests shown by AMD both had a mix of 1T and MT tests but it does seem like their Zen 5 test suite is a little bit heavier on the MT side than the prior two.

 

[Hitman928]

I question the reported clock rate.
And a further caveat: there are various subtests that could use AVX-512 and VNNI (if enabled by GB6) so comparing composite score of Phoenix to Strix may be questionable. Though I didn't see any stand outs in this comparison.

Definitely still need some actual controlled testing but these leaks are just enough so that the train has slowed to a calm pace rather than crash completely .

 

[Timorous]

Definitely still need some actual controlled testing but these leaks are just enough so that the train has slowed to a calm pace rather than crash completely .

Calm before the storm?

 

[ToTTenTranz]

Is it being choked by the limited cache?

Compute tests are usually the least bandwidth-demanding tests out there.

This should be the most favorable comparison of a 880M / 890M vs a 780M.

 

[adroc_thurston]

Is it being choked by the limited cache?

no, L2 is fine.
The lack of MALL will hurt relatively in gaming actual, which is why Halo has it.

This should be the most favorable comparison of a 880M / 890M vs a 780M.

The opposite, busto RDNA3 loves compute runs and hates drawing triangles.

 

[Mopetar]

It seems AMD used (for whatever reason) a mix of 1T and nT workloads for the "IPC" average. This is way different from previous Zen reveals. GB results seem to indicate that Zen 5 doesn't have as good SMT scaling as Zen 4. We'll have to wait for detailed benchmarks/reviews before we know for sure.

SMT scaling being lower would imply that they've removed some bottlenecks that previously stalled the hardware thread or otherwise left it unable to utilize the core to a greater extent.

 

[ToTTenTranz]

The opposite, busto RDNA3 loves compute runs and hates drawing triangles.

Erm they're both RDNA3 though?

 

[adroc_thurston]

Erm they're both RDNA3 though?

Sort of. 3.5 actually works.

 

[igor_kavinski]

Honestly, I am thinking that at this point, before we get any meaning to extra cores, we'll need a new pattern/language/compiler/profiler that allows automated or greatly simplified threading.

Someone should go ahead and create a game engine in Golang or Rust.

 

[adroc_thurston]

or Rust.

Embark people kinda tried?

 

[Mahboi]

Embark people kinda tried?

Someone should go ahead and create a game engine in Golang or Rust.

DO NOT​

REDACTED start.

 

[Saylick]

I haven't been tuning in as often over the last few days due to work, so what did I miss?

It seems that some people are getting whiplash from the Zen 5 hype train decelerating after Computex to suddenly accelerating??

 

[H433x0n]

It's wrong.(frequency)

ASUSTeK COMPUTER INC. ASUS TUF Gaming A14 FA401WV - Geekbench

Benchmark results for an ASUSTeK COMPUTER INC. ASUS TUF Gaming A14 FA401WV with an AMD Eng Sample: 100-000000994-37_Y processor.

browser.geekbench.com

View attachment 100723

This score isn’t too bad. So we’ve got scores ranging from 2550-2800 so far.

 

[Mahboi]

I haven't been tuning in as often over the last few days due to work, so what did I miss?

It seems that some people are getting whiplash from the Zen 5 hype train decelerating after Computex to suddenly accelerating??

Frankly not that much, things are still up in the air in a lot of ways.
We're not quiiiite sure about some oddities in the AMD presentation so the hopium is that it'll be slightly better than 16%, Intel promises much but may also disappoint, Novideo confirmed their name by having nothing to show.

At this point the best Computex content is Hairy Steve going around the stalls and 'miring all the pretty cases with way too many fans.